Skip to main content

Amazon Bedrock Monitoring in ObserveOps

Overview

The Amazon Bedrock integration with ObserveOps (formerly known as AIOps) collects operational telemetry from Amazon Bedrock, AWS's fully managed foundation model service. It monitors provisioned throughput utilization, guardrail invocations and interventions, inference profile states, and model invocation performance including latency, token consumption, and error rates.

These metrics help AI platform teams track model usage costs, detect throttling and errors, and ensure guardrails are operating as expected across generative AI workloads.

Prerequisites

  • Amazon Bedrock is enabled in the target AWS account and region.
  • The IAM role or user used for integration has read access to Bedrock resources (bedrock:ListProvisionedModelThroughputs, bedrock:GetProvisionedModelThroughput, bedrock:ListGuardrails, bedrock:GetGuardrail, cloudwatch:GetMetricData).
  • Required Bedrock and CloudWatch API endpoints are reachable from ObserveOps.
  • The AWS account is added in discovery with correct credentials and region configuration.

List of Supported KPIs

Provisioned Throughput

MetricDescriptionType
aws.bedrock.provisioned.stateCurrent state of the provisioned throughput configuration.String
aws.bedrock.provisioned.model.unitNumber of model units allocated in the provisioned throughput.Count
aws.bedrock.provisioned.nameName of the provisioned throughput configuration.String
aws.bedrock.provisioned.arnAmazon Resource Name of the provisioned throughput.String
aws.bedrock.provisioned.model.arnARN of the foundation model associated with provisioned throughput.String
aws.bedrock.provisioned.commitment.durationCommitment duration of the provisioned throughput agreement.String
aws.bedrock.provisioned.utilization.percentUtilization of the provisioned throughput capacity as a percentage.Percent

Guardrails

MetricDescriptionType
aws.bedrock.guardrailsTotal number of guardrails configured in the account.Count
aws.bedrock.guardrailIdentifier of an individual guardrail.String
aws.bedrock.guardrail.stateCurrent state of the guardrail.String
aws.bedrock.guardrail.versionVersion of the guardrail configuration.String
aws.bedrock.guardrail.invocationsTotal number of times the guardrail was invoked.Count
aws.bedrock.guardrail.interventionsNumber of times the guardrail intervened on a request.Count

Inference Profiles

MetricDescriptionType
aws.bedrock.inference.profileIdentifier of an inference profile.String
aws.bedrock.inference.profile.idUnique ID assigned to the inference profile.String
aws.bedrock.inference.profile.stateCurrent state of the inference profile.String

Invocation Performance

MetricDescriptionType
aws.bedrock.invocationsTotal number of model invocations.Count
aws.bedrock.invocation.client.errorsNumber of invocations that failed due to client-side errors.Count
aws.bedrock.invocation.server.errorsNumber of invocations that failed due to server-side errors.Count
aws.bedrock.invocation.throttlesNumber of invocations throttled due to rate limits.Count
aws.bedrock.invocation.latency.msAverage end-to-end latency of model invocations in milliseconds.Milliseconds
aws.bedrock.first.byte.latency.msTime from request submission to first response byte in milliseconds.Milliseconds
aws.bedrock.token.inputTotal number of input tokens processed across all invocations.Count
aws.bedrock.token.outputTotal number of output tokens generated across all invocations.Count
aws.bedrock.legacy.model.invocationsNumber of invocations using legacy foundation models.Count